A NNALES DE L ’I. H. P., SECTION C
R OBERT L. J ERRARD
H ALIL M ETE S ONER
Scaling limits and regularity results for a class of Ginzburg-Landau systems
Annales de l’I. H. P., section C, tome 16, n
o4 (1999), p. 423-466
<http://www.numdam.org/item?id=AIHPC_1999__16_4_423_0>
© Gauthier-Villars, 1999, tous droits réservés.
L’accès aux archives de la revue « Annales de l’I. H. P., section C » (http://www.elsevier.com/locate/anihpc) implique l’accord avec les condi- tions générales d’utilisation (http://www.numdam.org/conditions). Toute uti- lisation commerciale ou impression systématique est constitutive d’une infraction pénale. Toute copie ou impression de ce fichier doit conte- nir la présente mention de copyright.
Article numérisé dans le cadre du programme Numérisation de documents anciens mathématiques
http://www.numdam.org/
Scaling limits and regularity results
for
aclass of Ginzburg-Landau systems
Robert L. JERRARD *
Department of Mathematics University of Illinois 1409 West Green Street Urbana, IL 61801, USA
Halil Mete
SONER ~
Department of Mathematics Carnegie Mellon University Pittsburgh, PA 15213, USA Vol. 16, n° 4, 1999, p. 423-466 Analyse non linéaire
ABSTRACT. - We
study
a class ofparabolic systems
which includes theGinzburg-Landau
heat flowequation,
for uE :
R d R 2, as
well as some naturalquasilinear generalizations
forfunctions
taking
values in > 2.We prove that for solutions of the
general system,
thelimiting support
asE -~ 0 of the energy measure is a codimension k manifold which evolves via mean curvature.
We also establish some local
regularity results
which holduniformly
in E. In
particular,
we establish asmall-energy regulity
theorem for thegeneral system,
and we prove astronger regularity
result for the usualGinzburg-Landau equation
onR2.
©Elsevier, Paris
* Partially supported by the Army Research Office and the National Science Foundation through the Center for Nonlinear Analysis and by the NSF grant DMS-9200801.
t Partially supported by the Army Research Office and the National Science Foundation
through the Center for Nonlinear Analysis and by the NSF grants DMS-9200801, DMS-9500940, and by the ARO grant DAAH04-95-1-0226.
Annales de l’lnstitut Henri Poincaré - Analyse non linéaire - 0294-1449 Vol. 16/99/04/© Elsevier, Paris
424 R. L. JERRARD AND H. M. SONER
RESUME - Nous etudions une classe de
systemes paraboliques qui
comprennent 1’ equation
de chaleurGinzburg-Landau,
pour uE : Rd R2,
ainsi que desgeneralisations quasilineaires
pour desfonctions
prenant
leurs valeurs dans>
2.Nous
prouvons
que, pour les solutions dusysteme general,
lesupport
limite
(lorsque
E ~0)
de mesured’énergie
est une variete decodimension k qui
evolue selon sa courbure moyenne.Nous etablissons en addition
quelques
resultats deregularite locale, qui
sont valides uniformement en E. @
Elsevier,
Paris1.
INTRODUCTION
We
p resent
in this paper acollection
of resultsconcerning
theasymptotic regularity
andqualitative behavior
of solutions of theGinzburg-Landau
system,
We also
propose
andstudy
a class ofequations
which we believe arenatural generalizations
of(1.1).
Thesesystems
have the formHere
Of
special
interest is the case where p =k;
this is a directgeneralization
of
(1.1).
Annales de l’Institut Henri Poincaré - Analyse non linéaire
The
Ginzburg-Landau system
arises in avariety
of contexts,including
models of
superconductivity
and ofsystems
ofcoupled
oscillators near abifurcation
point,
see forexample
Kuramoto[27]. Recently
the associatedminimization
problem
has been studied ingreat
detailby Betheul, Brezis,
and Helein
[4], [5],
with refinementsby
Struwe[26],
among others.Neu
[19],
Pismen and Rubinstein[20],
Rubinstein[21],
E[9],
and othershave
analysed (1.1)
and the associatedSchroedinger-type equation using
matched
asymptotic expansions.
A number of results on the behavior of(1.1)
in two space dimensions were obtained
by
Lin[17], [18].
We view
(1.2)
as a naturalgeneralization
of(1.1)
toenergies
withnonquadratic growth
in thegradient
term. Given a solution uE of(1.2)
we define
We think of EE as a energy
density
for thegeneralized Ginzburg-Landau
system.
Thisinterpretation
is motivatedby
the fact thatis
formally
aLyapunov
functional for(1.2).
We remark that(1.2)
is notan
equation
forgradient
flow for the functional IE.However,
it retains many of the estimates satisfiedby (1.1),
estimates which are crucial to anyanalysis
ofproperties
of solutions.(These
estimates arechiefly presented
in Section
2).
Also,
in the same way that(1.1)
is a kind of modelproblem
forcodimension 2
pattern formation,
thegeneralized system (1.2)
can serve as a modelproblem
for thestudy
ofhigher
codimensionpattern
formation.This view is
supported by
the results wepresent
in Section3,
which are discussedimmediately
below.Our results fall into two classes.
First,
we characterize thequalitative
behavior of solutions of
(1.2)
in the limit as E -~0,
in the case where d> 1~
= p. Moreprecisely, given
afamily
of solutions uE of(1.2)
withappropriate
initialdata,
we define an associatedfamily
of measuresv;,
andwe show that the
support
of these measures, in thelimit,
formsexactly
a( d - k)-dimensional
submanifold which evolves via codimension k meancurvature
flow,
at least for short times.This
result,
whichoccupies
Section3,
confirms the formalcomputations
of Rubinstein
[21],
Pismen and Rubinstein[20],
and E[9]
for the usualGinzburg-Landau system (1.1)
in three spacedimensions,
and alsoapplies
Vol. 16, n° 4-1999.
426 R. L. JERRARD AND H. M. SONER
to more
general
situations. It isclosely
related to a number of recent resultsabout the
asymptotic
behavior of solutions of scalarGinzburg-Landau equations
and relatedequations.
Forexample,
Chen[7], Evans,
Soner andSouganidis [11],
Ilmanen[23],
and Soner[23]
have shown that solutions of the Allen-Cahnequation
in asingular
limit exhibit asharp
interfacewhich evolves via codimension 1 mean curvature flow. The latter three papers establish this result
globally
intime, using
various weak notions of evolution via mean curvature.Analagous
results have been established for moregeneral
scalar reaction-diffusionequations by Barles,
Soner andSouganidis [2]
and Jerrard[12],
among others.The
larger part
of this paper is devoted toestablishing
someregularity
theorems. We first prove a small energy
regularity
result. In Section 4we prove that if certain
weighted integrals
of the energydensity
EE aresufficiently small,
then EE is in fact bounded in some smallerregion.
This result is valid
uniformly
forparameter
values E E(0.1].
Ourproof
uses a
monotonicity
formula and a Bochnerinequality, following
ideasof Struwe
[24],
and Chen and Struwe[8].
Small energyregularity
anda
covering argument imply partial regularity results,
as in Chen and Struwe[8].
In the
special
case of the usualGinzburg-Landau equation
inR2
x[0, T],
we establish much
stronger regularity
results. We prove that ifintegrals
ofthe energy
density
are bounded in someregion,
then in fact the energy ispointwise
bounded in a smallerregion.
Thisresult,
which isagain
uniformin E, follows from the small energy
regularity
via ablowup argument (Section 6)
and aLiouville-type
theorem(Section 7).
Theblowup argument
is similar to one found in Struwe[25].
This latter
regularity
result is used in another paperby
theauthors, [14]
in which we
completely
characterize theasymptotic
behavior of solutions of(1.1)
in H x[o, T ~,
where H CR2
and T > 0. Thisresult,
which isvalid
only locally
intime, provides rigorous proof
of formal results of Neu[19],
E[9]
and others.The paper starts with a collection of estimates in Section 2.
One issue we do not address is the
solvability
of(1.2).
It is well- known that(1.1)
admits smoothsolutions;
this follows from the work ofLadyzhenskaya, Solonnikov,
and Uraltseva[16],
as is verified inBauman, Chen, Phillips,
andSternberg [3],
forexample.
Results of this sort are not so obvious in the case of thegeneralized system (1.2).
It is not difficult toconstruct some sort of weak solutions of
(1.2),
forexample by discretizing
in
time, solving implicitly
at each timestep,
andpassing
to limits. To establishregularity, however,
seems torequire
apriori
estimates.Annales de l ’lnstitut Henri Poincaré - Analyse non linéaire
Such estimates are not, in
general,
valid forquasilinear systems,
butthey normally
hold forsystems
for which there is some sort of energydensity
which is itself a subsolution of an
elliptic
orparabolic equation.
Thisis the case for
(1.2),
as is shown inProposition
2.1. It is therefore not unreasonable toexpect
that the same estimate holds for(1.2),
and thus that smooth solutions exist. In this paper,however,
we focus on other issuesand
simpy
assume the existence of smooth solutions.We will
always
assume that the initial data for(1.2)
satisfiesMultiplying (1.2) by uE
anddefining wE : - ~ 2,
we discoverThe maximum
principle
thussuggests
that any reasonable solution shouldsatisfy
for all
(x, t)
ERd
x[0, oo). Similarly,
estimates in Section 2imply
that awell-behaved solution should have the
property
thatBoth of these statements will
hold, roughly speaking,
aslong
as there isno influx of energy
from
= +00. It is not hard to see, forexample,
that a solution
produced by
theimplicit
time discretization described above will have theseproperties.
We therefore further assume that for initial data asdescribed,
our solutionssatisfy
both(1.6)
and(1.7).
To establishthese estimates a
priori
wouldrequire
a delicateanalysis
andmight
not bepossible,
as is shownby
theexample
of the heatequation.
NOTATION AND
PRELIMINARIES
We will use thefollowing
notationthroughout
this paper.Integers d
and k willalways
denote the dimensions of the domain and the range,respectively,
of themappings
we consider.eE(.)
andE~(.)
willalways
be as defined in(1.2)
and(1.3),
where thepower p in the latter definition is understood to be the same as that in the
Vol. 16, n° 4-1999.
428 R. L. JERRARD AND H. M. SONER
generalized system (1.2).
We willnormally
write eE instead ofeE (uE ),
whenno confusion can
result,
and likewise EE .We
employ
the summation conventionthroughout.
Roman indicesi, j, , ...
are
always
understood to run from 1 tod,
andgreek
indices a,~,
...run from 1 to k.
Exceptions
will be indicatedexplicitly.
A scalarproduct
between matrices is denotedby
A :B,
so that forexample
I .-
We also use the notation
We will
normally
omit thesuperscript n
which indicates the dimension of the ambient space,displaying
itonly
when the dimension is not obvious from the context.Observe that if uE solves
(1.2)
for agiven
value of theparameter
E, thent)
.-uE(ax,
solves(1.2)
with e .-Similarly,
wehave
t)
_a2t). Rescaling
in thisfashion,
we canconvert statements about solutions of
(1.2)
forarbitrary
E into statementsabout solutions with E =
1,
forexample.
Whenever a statement of a theoremis invariant under this
rescaling,
itclearly
suffices to prove it for asingle
value of the
parameter
E. We will invoke this sort ofargument
from time to timeby saying,
without furtherexplanation,
that it suffices"by
arescaling argument"
to consider a certain case.2. ESTIMATES
In this section we collect some estimates that we will use
throughout
this paper.
We assume that uE is a smooth solution of
(1.2)
onRd
x[0,oo)
andthat
EE(~, 0)
EL1 (R d).
Following
asuggestion
of M. Grillakis we defineThe
following
fundamental identities are immediate consequences of theequation (1.2).
We haveAnnales de l’lnstitut Henri Poincaré - Analyse non linéaire
Given a smooth test function yy E x
~0, oo~ ),
wemultiply
the firstequation
aboveby ~
and the secondby Vyy,
then subtract to obtainWe
integrate
to findBy adding,
rather thansubtracting, equations (2.1)
and(2.2),
we obtainin a similar fashion
The
integration by parts
that we have carried out above isjustified
ifThe former follows from our
standing assumption (1.7). Invoking
the sameassumption,
the latter holds fora.e. t,
sinceand the
right-hand
side is finite a.e. t. Whenever weapply
the aboveestimates,
we willintegrate
them over some timeinterval,
so we cansafely ignore
the set of measure zero on whichpE ( ~, t)
is notintegrable.
We next show that the energy
density
EE solves a certainparabolic equation.
In the statement andproof
of this lemma we omit allsuperscripts
E,and we write e to mean
e(u)
=eE(uE).
PROPOSITION 2.1. - The energy
density
Esatisfies
Vol. 16, n ° 4-1999.
430 R. L. JERRARD AND H. M. SONER
Also,
Proof -
From the definition of E wecompute
We now
replace
ut and~ut
in the aboveequation by expressions
we obtainfrom the
generalized Ginzburg-Landau system (1.2), thereby obtaining
(We
have written outexplicitly
the terms for which there is some chance that more condensed notationmight
beambiguous.)
We also havefrom which we deduce that
From these we
obtain,
aftercancelling
several terms andcombining
termsof the same
form,
Annales de l ’lnstitut Henri Poincaré - Analyse non linéaire
From the
definition
of e we see thatThe above two
equations immediately imply
that(2.5)
holds.To
prove (2.6)
from(2.5),
note thatCauchy’s inequality gives
If
lul2
>1/2
then the first term in theright-hand
side isnegative. If,
onthe other hand, lul2 1/2
then(1 - ~u~2)2 4(1 - Therefore
With (2.5) this immediately yields (2.6).
DFinally
wederive
some L-bounds
for the energy. As thesebounds
depend
on ., weagain
indicateexplicitly
theparameter .
in what follows.From
(2.5)
weeasily
see that :=satisfies
~-~
Thus the
maximum principle implies
that for anysmooth solution
use and for alls,t
>0,
If we
strengthen
ourassumptions
on theinitial data,
weobtain
thefollowing
more
useful
result.PROPOSITION 2.2. -
Let u~
6C-(R’
xsmooth solution
of (1.2 )
with p >2, such
thatThe
conclusion
of the lemma followseasily
fromstandard regularity theory
if p = 2.Proof -
1.By rescaling
itsuffices
toconsider
the case e= 1.
Vol. 16, n° 4-1999.
432 R. L. JERRARD AND H. M. SONER
Let w :=
lul2 and 03C8
:= E +K(w - 1),
where K > 03BA will be fixedbelow. For a smooth
function §
letThen
using (1.5)
wecompute
thatThis with
(2.5) gives
where
2. We estimate
Hence
Annales de l ’lnstitut Henri Poincaré - Analyse non linéaire
on the set
{eP/2
>K(p - 2)}. Combining
thesecalculations,
we obtainNote that
if 03C8
> 2K then E > 2K and thuseP/2
>K(p - 2).
3. For p >
2,
set ..There is a
K(p)
> 1 such thatC(K, p)
> 2K + 1 for all K >K(p).
Moreover,
if K >K(p) and 03C8
2K +1,
thenTherefore
by taking
K =K(p)
v x in the definition of~,
weget
4. If we
set § :_ ~
V2K,
then.C~
0 on~~
2K +1} (in
the senseof
viscosity solutions)
and~(x, 0) =
2K. Letand define
From
(2.7)
we deduce thatc( . )
is continuous and thatto
> 0.Also, ,C~
0on
Rd
x( 0, to )
and so the maximumprinciple implies
that if tto
then~(x, t) ~(x, 0) =
2K. Thusto
= +ooand §
2I~ onR~
x[0, oo).
D3. CONVERGENCE TO CODIMENSION k MEAN CURVATURE FLOW
In this section we consider examine
asymptotic
behavior of solutions of thegeneralized Ginzburg-Landau system
in the case d > 1~ = p.For this purpose, it is convenient to introduce the normalized measure
Vol. 16, n 4-1999.
434 R. L. JERRARD AND H. M. SONER
In the
following,
we assume thatro
is a smooth embeddedcompact
(d - k)-dimensional
submanifold ofRd,
and that is a smooth codimension k mean curvature flowstarting
fromfa,
for some T > 0. We letT ~ Rd
x[0, T]
denote the setswept
outby f t,
i.e.Also,
we defineSince r is smooth and
compact,
we can find a number ~o > 0 and a smoothfunction ~
such thatand
Ambrosio and Soner
[ 1 establish
severalproperties
of thefunction 1 203B42
ina recent paper. Their results
immediately imply
that ri has thefollowing properties:
THEOREM 3.1. - For
(x, t)
in aneighborhood of 0393,
the matrixt)
has k
eigenvalues equal
to one, and eachof
theremaining
d - keigenvalues satisfies
theestimate t) ( Cb(x, t).
Inparticular, ~2r~(x, t)
is aprojection
onto a k-dimensionalsubspace
when(x, t)
E T.Moreover,
In
fact,
Ambrosio and Soner[1] ] show
that for a smoothevolving
manifoldas
above, t) gives
the normalvelocity
vector ofrt
at(x, t)
Er,
and
t) equals
the mean curvature vector. so the aboveequation precisely
characterizes smooth codimension k mean curvature flow.We will use these results in the
following
form:COROLLARY 3.1. - For
any 03BE
ERd
and all(x, t) in a neighborhood of
r we haveAnnales de l’lnstitut Henri Poincaré - Analyse non linéaire
Also,
risatisfies
Proof. -
The first assertion is immediate from the above characterization of theeigenvalues
of near F.To
verify
the secondassertion,
first note thatThe first of these
equalities
holdsbecause ~
attains its minimum onr,
andthe second follows from the
description
of in Theorem 3.1. If we let~
=A??,
we thus have(again using
Theorem3.1)
Given any
(x, t)
ERd
x[0, T],
we canfind y
Ert
such that =8(x, t).
We then have
The
following
theorem is an easy consequence of theseproperties
of r~.THEOREM 3.2. -
Suppose
thatRd -~ R~
is a smooth solutionof
the
generalized Ginzburg-Landau system (1.2)
with p = k and initial datahE(X) for
whichas E --~ 0. Then
Proof. -
We use the smoothfunction ~
defined above in theweighted
energy estimate
(2.3). Dropping
anegative
term andusing (3.2)
and thedefinition of we have
Vol. 16, n° 4-1999.
436 R. L. JERRARD AND H. M. SONER
Select s E
(0, ao]
such that(3.1)
holds on the setr:
=t) s~.
Thisnumber s may be chosen
uniformly
for t E[0, T),
so we may assume that~~~z~~~~~~1
C onR‘~ ~ Tt
for some constantC, uniformly
in[0, T] .
ThenMoreover,
The three
preceding inequalities together yield
Gronwall’s
inequality
nowimmediately implies
thatfor all 0 t T.
Dividing by log §,
we obtain the conclusion of thetheorem. D .
Remark. - This
proof
may be seen as aPohozaev-type estimate,
as usedfor
example
inBauman, Chen, Phillips,
andSternberg [3].
The
hypotheses
of the above theorem mentiononly
the initial distribution of energy. If we assume in addition that hE exhibits a vortex-like structurealong
cross-sections ofFo,
so thatFo
is a"topological defect",
then wecan strenthen the above result.
Because
ht
is assumed to be a smooth codimension kmanifold,
at eachy E
rt
we may find vectorsnl(y, t),..., nk(y, t)
ERd
such that each na is normal toTt,
andn(3
=8a(3.
We assume moreover thatr t
is
orientable,
so that(y, t)
~--~na (y, t)
may be taken to be smooth andglobally
well-defined on r.For y
Ert,
we definewE ( ~; ~, t) : R~ --~ R~ by
Annales de l’Institut Henri Poincaré - Analyse non linéaire
THEOREM 3.3. -
Suppose
that uE is a smooth solutionof (1.2)
with p = kand initial data uE
(x, 0)
= hE(x) satisfying
uniformly for
E > 0.If
in addition there exists x, ~1 > 0 such thatK/e
andfor all y
E E thenfor all ~
ECo (Rd).
The constant comes from Lemma 3.2 below.
Remark. - Theorems 3.2 and 3.3 taken
together imply
that if vt is any weak limit ofv;,
then thesupport of vt exactly equals rt .
As the
sign
ofdeg(wE ( . ;
y,t) ) depends
of the choice of ...,n k,
wemay take it to be
positive,
without any loss ofgenerality.
We assume a > 0 is fixed and we introduce
the
notationWe denote
typical points
inB4u
and K as xand y respectively.
Given afunction vE : U ~
R/B
we further defineHere
leb1
denotes 1-dimensionallebesgue
measure. We may think ofYtE
as the subset of
points
in K at which the cross-section at time t exhibitsan isolated vortex, in a weak sense.
The
following
two estimates are proven in Jerrard[13]
as Theorem 6.1 and Theorem 4.2respectively.
Vol. 16, n° 4-1999.
438 R. L. JERRARD AND H. M. SONER
for
some x >0,
and assume thatfor all y
EK,
r E[03C3, 403C3],
and thatfor
all t E~0, T].
Thenfor
all t E[0, T ~ .
LEMMA 3.2. -
Suppose
that~E
E and thatand that
Then
The constant
K( k)
isgiven explicitly
in Jerrard[13].
Using
these wepresent
theProof of
Theorem 3.3.1. For cr E
(o, ~1]
to bedetermined,
define U as above.First we define a
map 03C8
ECX> (K
x[0, T]; r)
such that for every t E[o, T], t)
is adiffeomorphism
of K onto a subset ofTt.
Now wedefine 03A8 : U x
[0, T]
~Rd by
Note that for fixed
(y, t), W(.,
y,t)
mapsR~
onto the normal space toAnnales de l ’lnstitut Henri Poincaré - Analyse non linéaire
We may assume that
I
C and that theC-1
>0,
where= det is the Jacobian of ~.
We also assume that a is small
enough
that ~ is one-to-one.Finally
we definevE(x, y, t)
.-2. We now
verify
that vE satisfies thehypotheses
of Lemma 3.1 Since uE is assumedsmooth,
it is evident that themap t H vE { ~, ., t)
is continuousin the norm of
W 1 ~ °° .
We nextcompute
Also,
it is clear thatso the condition on the
degree
ofv~ ( ~, ~, 0)
followsimmediately
from(3.5)
and our choice of cr.
Finally,
note that and soThe final
equality
follows from(3.4) by
the calculation in theproof
ofTheorem 3.2.
3. Lemma 3.1 therefore asserts that
(3.8)
holds.We now
define, for y E rt
It is clear that we can find a finite collection of of the form described above such that
Vol. 16, n ° 4-1999.
440 R. L. JERRARD AND H. M. SONER
Thus
(3.8)
and(3.9) imply
that4. For x
sufficiently
close toFt,
letp(x)
Ert
be theunique point
ofrt satisfying
Fix r
so
small thatp(x)
is well-defined on{~(-,~) 4o-}.
Note that forY 6
~,
This is an immediate consequence of Lemma 3.2.
In the calculations
below, Jp
denotes the Jacobian of p,Jp
:=[det
Heredp
denotes thegradient
of p considered as a map fromRd
into and thus isexpressed
as a(d - k)
x dmatrix,
afterchoosing
bases for therespective tangent
spaces. Inparticular,
with this definition the
change
of variables that weemploy
below is valid.For every
smooth, compactly supported §
we havewhere
and, by
a version of the co-areaformula,
In the last
step
we have used(3.11).
Annales de l’Institut Henri Poincaré - Analyse non linéaire
5. Fix any
subsequence
and a measure v such that --~ v.By
Theorem
3.2,
we know thatspt v
Cft.
It follows thatWe will show in Lemma 3.3 below that
Jp(y) =
1 for y Eft.
ThusIi
vanishes as E ~ 0.
Also, (3.12)
and(3.10) evidently imply
thatLEMMA 3.3. - For p as
defined
aboveand y
EFt, Jp(y)
= 1.Proof. - Fix y ~ 0393t
and orthonormal vectors To , ... , Td-k which spanTy rt . Taking
the standard basis e 1, ... , ed as a basis forTy Rd,
the matrixdp
has the formAfter a
relabelling
we may assume that ei == Ti for i =1,...,
d - k andthat are normal to
ft
at y.We claim that
Indeed,
for anyi = 1, ... , d - ~, by
the definition of p,since V6 is normal to
rt.
Sincep(y)
= y, thisimplies
thatwhich
implies (3.13).
Also, for j
> d - k and hsufficiently small,
similarreasoning
showsthat
p(y
+hej)
=p(y).
Thus(dp)ij
= 0whenever j
> d - k. With(3.13)
and the definition of
Jp,
thisimplies
the conclusion of the lemma. DVol. 16, n° 4-1999.
442 R. L. JERRARD AND H. M. SONER
In the remainder of this
section,
webriefly
indicate a way to constructinitial data hE for
(1.2)
in such a way that theresulting solutions,
ifsmooth,
will
satisfy
thehypotheses
of Theorems 3.2 and 3.3.We
impose
sometopological
restrictions onro by assuming
that thereexist
smooth, bounded,
open sets =l,
...,k,
such thatFor a =
1,..., ~,
letda
be the(signed)
distance to so thatSince the sets
C~a
are assumedsmooth,
each functionda
is smoothnear We assume in addition to
(3.14)
thaton ro.
For ex =
1,..., k,
let da be smooth functions such thatLet d :
R~
be the vector-valued function whose athcomponent
is da . Note that d is related to theordinary
distance functionb( ~, 0),
definedabove, by
Finally,
note thatassumption (3.14) implies
that -k-1/2(1, ..., 1)
as
]
-~ oo, so we may find dsatisfying
the aboveconditions,
for whichthere exists some number M such that
Remark. - 1.
Assumption (3.14)
appears to be a necessary condition for the existence of initial data with therequired properties.
Given(3.17),
onecan
modify
the setsOa locally
nearro
to arrange that(3.15)
be satisfied.Annales de l’Institut Henri Poincaré - Analyse non linéaire
2.
Assumption (3.14)
is satisfiedby
anyFo
which can be embedded as a codimension 1 manifold inAlso,
it isclearly preserved
underhomotopy.
Let v :
Rk R~
be a function of the formv(x)
= for ascalar function p such that
Then for
e(v)
+W(v),
we haveWe define
One can then
verify
thatin the sense of
distributions,
and it is clear that(3.5)
holds.Moreover,
onecan
verify
that hE satisfies thehypotheses
ofProposition 2.2,
and thus thata smooth solution uE with hE as initial data satisfies
So hE has all of the desired
properties.
4. SMALL ENERGY REGULARITY
In this section we establish a small energy
regularity
theorem for solutions of thegeneralized Ginzburg-Landau system.
The basicargument
we follow was introduced
by
Schoen[22]
forstationary
harmonic maps andgeneralized by
Struwe[24]
and Chen and Struwe[8]
to the case ofheat flow for harmonic maps and for
Ginzburg-Landau type approximations
of harmonic maps.
The
proof
relies on amonotonicity
lemma and aBochner-type inequality,
that
is,
a differentialinequality
which is satisfiedby
the energy. The mainVol. 16, n° 4-1999.
444 R. L. JERRARD AND H. M. SONER
novelty
here is the observation that these estimates are available in thismore
general
context, as well as the fact that our result is local in nature. Inproblems involving asymptotic
behavior of solutions ofGinzburg-Landau
type systems, global
energyestimates, independent
of E,typically
do nothold. Thus the local character of our estimates is very useful in these
applications.
Small energy
regularity
results of the sort that we establish here canbe used with
covering arguments
to deducepartial regularity results,
as inChen and Struwe
[8].
We start
by establishing
amonotonicity formula,
which weget by putting
an
appropriate
testfunction ~
in theidentity (2.5).
We first define this function:Let f : [0, oo)
-~[0,1]
be a smoothnonincreasing
function such thatAlso, define p : Rd
x(0, oo)
- Rby
where I is the
identity
matrix. We then haveAnnales de l ’lnstitut Henri Poincaré - Analyse non linéaire
So with this choice of 7y in
(2.4)
we obtain(using
the factthat ~
isnonnegative)
By
the definition of ~y, theintegrals
on theright-hand
side above aresupported
on{x : 1~4 ~x~ 1~2}. Recalling
that~~~y~2/y
we have
Also,
forIxl
>1/4,
Thus we have established the
following
localmonotonicity
formula.LEMMA 4.1. - The measures
satisfy
the estimateBefore
stating
oursmall-energy regularity result,
we introduce somenotation. For Xo E
Rd,
r >0,
and 0 tto,
letwhere q
and p are defined at thebeginning
of this section. We write at to meanai.
Note that thear
is scale-invariant in thefollowing
Vol. 16, nO 4-1999.
446 R. L. JERRARD AND H. M. SONER
sense: Given a function uE
solving (1.2),
we may define a rescaled functionby ic(x, t)
=uE(xo
+Rx, R2t).
We also define E =E(x, t)
=2/p(eE(u))P~2.
As remarked in theintroduction,
ic solves thesystem (1.2)
withscaling E,
andThus, using
the fact thatp(Ry, s)
= we obtainby
achange
of variablesIn
particular, by taking R
= r we can convert statements aboutar
tostatements about aE .
We now
change notation, using ~
to denote a small constant which will be chosen below. We also introduce the notationWe now have
THEOREM 4.1
(local small-energy regularity). - Suppose
that uE is asolution
of (1.2)
onBr
x[To,
with E r.Suppose
also that there exists ~ > 0 suchthat for
all x EBr,
s r with CBr,
and allt E
[To,
we haveThen there are
positive
constants ri, po, and C such thatif
for
some(xo, to)
EBr /4
x[To
+T2, Ti]
and T E(0,
then we haveAnnales de l’Institut Henri Poincaré - Analyse non linéaire
Proof. -
1. We first claim that it suffices to establish the theorem under theassumption
that T =r vfij. Indeed,
ifT2
then we define rby insisting
that T =
f vfij.
Note that r r, and so7(~x~~r) y(~:c~)
for all x. ThusClearly
also(4.2)
continues to hold if r isreplaced by
r. We may then use r instead of r in theproof,
and the desiredequality
will be satisfied.Next, by rescaling
we may set r = 1. Thus we assume thatThe
constant ~
E(0,1]
will be fixed at the end of theproof.
After thesenormalizations, (4.2) implies
that2. For 03C10 ~
(0,1/4]
to bechosen,
letThus
E (x, t)
2P[ in Estimate(2.6)
nowimplies
thatVol. 16, nO 4-1999.
448 R. L. JERRARD AND H. M. SONER
Note also that it now suffices to show that C for
appropriate
choices of q, po.
Indeed,
if we have thisestimate,
thenwhich is the conclusion of the theorem.
Then
and w in
P~ (x, I).
The coefficients in the aboveequation satisfy
so
by
aparabolic
Harnackinequality
fornondivergence
structureequations,
see
Krylov
and Safonov[15],
whichdepends only
on the above boundson the
coefficients,
we have4.
Since ~7
1 and po1/4,
and so for
(x, t)
E~a (~, t),
we haveThus
by
themonotonicity formula,
Lemma 4.1.Annales de l’lnstitut Henri Poincaré - Analyse non linéaire
Recall that
by construction, t to,
so the last term on theright
handside above is bounded
by
5.
By translation,
we may set xo =0,
and wedefine i := t
+~2 - to.
Observe that
by
construction we haveWe now claim that if po is
sufficiently small,
thenWe write
where
Recalling that ==
1 onB1/4(0),
we haveif
1/8
>H.
which may beachieved,
forgiven
q,by adjusting
po.Taking
po still smalleryields Ii
To estimate note that if
p(x -
+i) -
+i)
> 0 thenWe rewrite this
inequality
asVol. 16, n ° 4-1999.
450 R. L. JERRARD AND H. M. SONER
This
implies
thatusing (4.6).
Thusif po is chosen small
enough.
With(4.3)
thisimplies
thatI2 r~~3.
The estimate of
13
is very similar to that ofI2,
so we omit it. Thuswe have proven our claim.
6.
Putting together steps
4 and5,
we find thatTaking ~
small nowgives
As remarked in
step 2,
thisimmediately yields
the conclusion of the theorem. D5. SOME VARIATIONS
By modifying
theargument
of the small energyregularity theorem,
weobtain
slightly
different results which will be useful later on.PROPOSITION 5.1. -
Suppose
that u~ solves(l.l )
onBr
x whereE r. Assume also that there exists ~ > 0 such that
Then there exist constants
C ( ~ ) , T()
such thatRemark. - Note that this
applies only
to the usualGinzburg-Landau
system
withquadratic growth.
Annales de l ’lnstitut Henri Poincaré - Analyse non linéaire
Proof. -
1. As in theproof
ofProposition 4.1,
it suffices to prove theresult for r = 1 > E. We may also assume
by
a translation thatTo
= 0.Take Xo E
B1 /2, to 1 / 16
to be fixedlater,
and defineExactly
as in theproof
ofProposition
4.1 we select ao E(0, to)
and(x, t)
EPao
such thatWe further define 3 :=
( to - Q~)/2. Following
theargument
ofsteps
1-4of the
proof
ofProposition 4.1,
we find thatfor some 3 3.
2.
By
themonotonicity lemma,
the definition of aE and the assumed L°° bounds onE~ ( ~, 0),
From the definitions we
have t + jj2 2to,
so withStep
1 we obtainSo there exists some T > 0 such that if
to
T, thenNext define
Vol. 16, n° 4-1999.